khabib_sketch_LoRA

Khabib Sketch SDXL LoRA

A LoRA adaptation of SDXL to produce sketches of the MMA fighter and G.O.A.T Khabib.

<figure> <img src="https://i.imgur.com/eIn5oqJ.png" alt="Khabib" width="256" height="256"> <figcaption>Sketch of Khabib fighting a Bengal Tiger</figcaption> </figure>

These are LoRA adaption weights for stabilityai/stable-diffusion-xl-base-1.0. The weights were trained on sketches of Khabib by ritwikraha using DreamBooth. You can find some example images in the following.

Special VAE used for training: madebyollin/sdxl-vae-fp16-fix.

DataSet: custom hand-drawn sketches by ritwikraha

Usage

!pip install diffusers accelerate -q
import torch
from PIL import Image
from diffusers import DiffusionPipeline, AutoencoderKL

vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    vae=vae,
    torch_dtype=torch.float16,
    variant="fp16",
    use_safetensors=True
)
pipe.load_lora_weights('ritwikraha/khabib_sketch_LoRA')
_ = pipe.to("cuda")

prompt = "a sketch of TOK khabib pointing at another khabib like the spiderman meme, monchrome, pen sketch"
negative_prompt ="ugly face, multiple bodies, bad anatomy, disfigured, extra fingers"
image = pipe(prompt=prompt,
             negative_prompt=negative_prompt,
             guidance_scale=3,
             num_inference_steps=50).images[0]
image

Examples

Tips

• The examples are all sketches created in Procreate so prompts with words like sketch, and monochrome work best
• Use a negative prompt and guidance scale for the model
• Images at 1024X1024 will be better than other dimensions